Review on classification based on artificial

International Journal of Ambient Systems and Applications (IJASA) Vol.2, No.4, December 2014
DOI:10.5121/ijasa.2014.2402 11
REVIEW ON CLASSIFICATION BASED ON ARTIFICIAL
NEURAL NETWORKS
Saravanan K1
and S. Sasithra2
1
Assistant Professor/Department of Computer Engineering, Erode Sengunthar
Engineering College, Erode, India
2
Department of Computer Engineering, Erode Sengunthar Engineering College, Erode,
India
ABSTRACT
Artificial neural networks (ANN) consider classification as one of the most dynamic research and
application areas. ANN is the branch of Artificial Intelligence (AI). The neural network was trained by
back propagation algorithm. The different combinations of functions and its effect while using ANN as a
classifier is studied and the correctness of these functions are analyzed for various kinds of datasets. The
back propagation neural network (BPNN) can be used as a highly successful tool for dataset classification
with suitable combination of training, learning and transfer functions. When the maximum likelihood
method was compared with backpropagation neural network method, the BPNN was more accurate than
maximum likelihood method. A high predictive ability with stable and well functioning BPNN is possible.
Multilayer feed-forward neural network algorithm is also used for classification. However BPNN proves to
be more effective than other classification algorithms.
KEYWORDS
Back propagation algorithm, Maximum Likelihood method, Multilayer feed-forward neural network.
1. INTRODUCTION
A neural network model which is the branch of artificial intelligence is generally referred to as
artificial neural networks (ANNs). ANN teaches the system to execute task, instead of
programming computational system to do definite tasks. To perform such tasks, Artificial
Intelligence System (AI) is generated. It is a pragmatic model which can quickly and precisely
find the patterns buried in data that replicate useful knowledge. One case of these AI models is
neural networks. AI systems should discover from data on a constant basis. In the areas of
medical diagnosis relationships with dissimilar data, the most available techniques are the
Artificial Intelligence techniques.
An artificial neural network is made up of many artificial neurons which are correlated together in
accordance with explicit network architecture. The objective of the neural network is to convert
the inputs into significant outputs.The teaching mode can be supervised or unsupervised. Neural
Networks learn in the presence of noise.
ANNs found their usage in many areas suc as,
• Bankruptcy prediction

12
• Speech recognition
• Product inspection
• Fault detection
Fig1. Artificial Neural Network
2. EXISTING SYSTEM
2.1. Back propagation Neural Network
Artificial neural networks (ANN) consider classification as one of the most dynamic research and
application areas. The major disadvantage in using ANN is to find the most appropriate grouping
of training, learning and transfer function for classifying the data sets with growing number of
features and classified sets. The different combinations of functions and its effect while using
ANN as a classifier is studied and the correctness of these functions are analyzed for various
kinds of datasets.
The real world problems which are represented by multidimensional datasets are taken from
medical background. The classification and clustering of these data sets are significant. The data
set is divided into training set and testing set and it has no usage in the training process. The
results are produced with the help of these datasets and it is used for testing. The training set is
taken from 2/3rd
of the dataset and the remaining has been taken as test set. This is made through
the assessment of the accuracy achieved through testing against these data sets. Then the network
is simulated with the same data.
The back propagation algorithm trains the neural network. Gradient descent method (GDM) was
used to decrease the mean squared error between network output and the actual error rate.The
following parameters are considered to measure the efficiency of the network,
• Rate of convergence
• No of epochs taken to converge the network.
• The calculated Mean Square Error (MSE).
With the appropriate combination of training, learning and transfer functions the dataset
classification uses the most successful tool called back propagation neural network. The
combination of TRAINLM, LEARNGDM and LOGSIG mentioned in the proposed work works

13
better for relatively smaller datasets but the combination TRAINSCG LEARNGDM and
LOGSIG is successful for larger datasets.
E. Hosseini Aria, J. Amini, M.R.Saradjian(2009) proposed a method for Classifying IRS-1D
Satellite Images[4].
The fitness of Back Propagation Neural Network (BPNN) for classification of remote sensing
images based on three steps is proposed. As an initial step, from the measures of first order
histogram measures the features are extracted.In the second step, feature classification based on
BPNN is done, and in the third step the outcomes are compared with the maximum likelihood
classification (MLC) method. The statistical features in this paper depend on the first-order
distribution measure such as mean, standard-deviation, skewness, kurtosis, energy, and entropy.
The network contains 3 layers. The input layer is fed with extracted features which contains 18
neurons.in the classification of IRS-1D satellite images six classes were used and the back
ropagation neural network was trained on these classes. The whole image was classified using
this trained network. The regions of Iran are taken for testing.
The IRS-1D satellite images uses Artificial neural network for the classification of images. The
major problem with the classification of IRS data is to choose a better method for training.
TrainLM method has been implemented on using back propagation neural networks algorithm on
IRS images.
There are various training algorithms for feed forward networks The gradient of the performance
function is used by all the algorithms to find out how to fiddle with the weights to decrease the
performance. The back propagation technique determines the gradient. This gradient performs
computational backwards through the network. When the maximum likelihood method was
compared with backpropagation neural network method, the BPNN was more accurate than
maximum likelihood method.The overall preciseness in MLC method is 75.00% whereas in
BPNN method is 85.19%.
Helena Grip, Fredrik(2003) proposed Whiplash-Associated Disorders by classifying of neck
movement patterns[5].
A novel method for the classification of neck movement patterns related to Whiplash-associated
disorders (WAD) using a flexible back propagation neural network (BPNN) is studied. WAD is a
common diagnosis after neck trauma, mainly caused by rear-ends car accidents. Since physical
injuries cannot be detected with the current imaging techniques, the diagnosis can be complex to
make. The dynamic range of the neck is often visually detected in patients with neck pain, but this
is a biased measure, and a more intentional decision support system, that gives a consistent and
more complete analysis of neck movement pattern, is needed. The estimation of the prognostic
ability of a BPNN, using neck movement variables as input is the main objective of the paper.
The collection of three-dimensional (3-D) neck movement data from 59 subjects with WAD and
56 control subjects is made with a proReflex system. Rotation angle and angular velocity were
measured using the direct helical axis method and motion variables and the results are extracted.
To increase the performance of BPNN a principal component analysis was performed which
reduces data. BPNNs with six hidden nodes had a yield of 0.89, a sensitivity of 0.90 and a
specificity of 0.88, which are very hopeful results. The results were predicted from the neck
movement analysis. The result was combined with a neural network where the origin of decision
support system is constructed, which classifies the suspected WAD.

14
The flexible back propagation neural network (BPNN) resulted in a correct calculation for 84
percent of the control subjects and 89 percent of the WAD, showing that a BPNN could be
appropriate for predicting motion characteristics. The presented method is very hopeful as an aid
to determine whether a patient with suspected WAD has a neck movement pattern that deviates
from that found in control subjects. A few perceptive variables seems to amplify the efficiency on
abbreviating the results . A high predictive ability with stable and well functioning BPNN is
presented using early stopping method.
Yu-guo Wang, Hua-peng Li(2010) proposed classification of Remote sensing images using
artificial neural network[6].
Artificial neural network (ANN) is a significant part of artificial intelligence, It has been
extensively used in the research field of remote sensing classification. The wetlands remote
sensing classification based on ANN is complicated, because of the intricate feature of wetlands
areas. The remote sensing image supervised classification is carried out on the training samples.
The clarity is examined and it was found that it is hard to guarantee because it will have an effect
on the classification results.This article proposed a method for sample purification to filter the
training samples based on statistical analysis theory for enhanced wetlands remote sensing
classification based on ANN. The BP ANN with a nonlinear mapping function gives better
classification results for intricate areas. The TM image of Honghe Wetlands National Nature
Reserve is chosen for classification. First, the statistical analysis theory to eradicate noise in
training samples is used. Then the original samples and purified samples are used to train the BP
ANN individually and created two classification maps of TM image based on two trained BP
ANN. At last, the classification accuracy between the two maps is compared. The statistical
analysis method for purifying training samples for remote sensing classification based on BP
ANN is performed. The experiments showed that it was an efficient method to develop image
classification.
Guoqiang Peter Zhang(2000) proposed the use of harmony search and back propagation
based ANN to classify breast cancer data.
Breast cancer is one of the most commonly seen cancer types. Analysis of breast cells
characteristics has great importance in diagnosis, treatment and following of this disease.
Through two different types of ANN algorithm 699 instances of breast cancer data which is
present in UCI are classified. In this study, abundant algorithms are used for training of artificial
neural network. In this study, two algorithms namely harmony search and back propagation
algorithms are used to train feed forward artificial neural network. While classifying the
performance values of classification are found by means of Accuracy/SSE/Regression
parameters. The performance values of back propagation are achieved as 94.1/0.007/0.92 whereas
the results obtained by the harmony search algorithm are 97.57/0.005/0.96 respectively. With the
help of harmony search based ANN algorithm breast cancer data are classified for the first time
with this study.
2.2 MULTILAYER FEEDFORWARD NETWORK
Guoqiang Peter Zhang(2000) proposed Neural Networks for Classification[7].
Classification is one of the most dynamic exploratory and application areas of neural network.
The issues of posterior probability estimation, link between neural and conventional classifiers,
learning and generalization tradeoff in classification, feature variable selection, effect of
misclassification costs are studied.

15
Multilayer feed-forward network is used for classification. Although many procedures are
available in traditional statistical classification, the usefulness depends on assumptions and
circumstances. Inorder to overcome the above disadvantage, neural networks are used since they
are,
• self-adaptive methods,
• universal functional approximators and
• non-linear models.
Many problems still stay unsettled even though neural networks have grown rapidly. As
represented prior, many researches should be dedicated to develop more successful and proficient
methods in neural model identification, feature variable selection, classifier combination, and
uneven misclassification treatment. As a practical decision making tool, systematic evaluation
and comparison of neural networks with other traditional classifiers need to be done. The lacks of
comparisons between other classifying methods and neural network have been proposed by some
authors in the existing literature. Due to the lack of comparisons, mixed results are produced and
reported in pragmatic studies.
There are other topics which are related to neural classification. The topics include sample size
issues, Bayesian analysis, wavelet network and network training model design and selection.
Some applications of neural network have been reviewed formerly and the problems mentioned
above are general to all applications. Abundant research opportunities are available in neural
classifiers. More research actions are generated by neural network classification because of its
multidisciplinary nature and brings about more productive outcomes in the future.
Abid Ali, Olaf Magnor and Matthias Schultalbers proposed Neural Network Based Pattern
Recognition Technique which detects the misfire in gasoline engines[9]
The application of ANN to misfire detection in gasoline engines is studied. A feed-forward
multiple layer neural network is used for the classification of firing and misfiring events.
Importance is given on the transaction between performance, computational cost and implement
ability and the system is applied on a production electronic control unit (ECU). The technique is
used to identify misfire events over the whole range of operation defined by official on-board
diagnosis (OBD) regulations and it is applied to a six-cylinder gasoline engine. Experimental
results on a passenger car are presented.
Misfire occurs because of some error in the combustion process. This error occurs because of
some insufficiency in ignition, injection or compression system of the engine or due to a
substandard fuel. In addition to their disagreeable effects on the drive quality, misfires in gasoline
engines amplify pollutants in exhaust gases and causes harm to the components. Misfire due to
some defect in ignition system will enlarge unburned mixture of hydrocarbons and oxygen in the
raw exhausts causing an increased exothermic reaction in the catalyst and increased HC and CO
emissions.
The investigation of realistic application of a neural network based pattern recognition technique
to misfire detection in gasoline engines is done. The difficulty of misfire detection is framed as a
pattern recognition problem. The classification of firing and misfiring events is done with the help
of feed-forward multiple-layer neural network. To detect the misfire events in six-cylinder
gasoline engines, the developed technique is applied over a wide range of operation. The results

16
are produced in this section. The results show that the developed technique detects all patters of
misfire.
R. Furferi, L. Governi and Y. Volpe(2011) proposed the classification of car seat fabrics[10].
Car seat fabrics are exclusively shaped textiles. A first-rate car seat fabric needs a little quantity
of spots and discolored areas. There are large and low number of spots in the fabrics which
characterizes them. The above mentioned fabric types are considered to be of poor quality. Most
of them is branded by a sponged-like appearance, characterized by spots and slightly discolored
areas. The human experts with the help of manual inspection and classification perform sponged-
like fabric grading. Though this manual classification and inspection proves to be valuable in
fabric grading, the procedure is skewed and its results may differ depending on the operator skills.
The computer based tool which classifies sponged-like fabrics similar to the classification
performed by skilled operators is shown here. Such a tool which is developed by an oppositely
devised machine vision system, is able to extract a lot of numerical parameters characterizing the
fabric veins and discolored areas. Since the Artificial Neural Network has to be trained, the
parameters are used which classifies the fabrics based on quality. At last, a relationship between
the ANN-based classification and the classification provided by fabric inspectors is done. The
validation set composed by 65 sponged-like fabrics is tested and it shows that it is able to classify
the fabrics into the correct quality class in 93.8% of the cases, similar to the selection provided by
human operators.
With the aim of training a ANN able to correctly classify the fabrics, a series of parameters
describing their appearance have been extracted by means of image processing-based algorithms;
such data are used as a training, validation and testing set for the ANN. In detail, as described
below, three kinds of parameters are extracted: colorimetric data, entropy curve and area
subtended to the entropy curve.
Most of the methods evaluates the classification performance by defining a dimensionless
parameter. A comparison between those methods and the proposed method was provided. For
instance, the correlation between measured and forecasted classifications of colored textiles is
stated to be in the range 0.85 – 0.98. The classification error of textured objects is less than 10%.
These results are comparable with the ones provided by the present work, thus allowing to state
that a satisfactory performance has been obtained.
3. PROBLEMS IN THE EXISTING SYSTEM
The major disadvantage in using ANN is to find the most appropriate grouping of training,
learning and transfer function for classifying the data sets with growing number of features and
classified sets. The classification of IRS-1D satellite images is done using Artificial Neural
Networks. The algorithm is used to categorize the imagery. The major problem with the
classification of IRS data is to choose a better method for training. Abundant algorithms are used
for training of artificial neural network. . Although many procedures are available in traditional
statistical classification, the usefulness depends on assumptions and circumstances. In order to
overcome the above disadvantage, neural networks are used.

17
Table1 Comparison of various classification methods based on artificial neural network
SNO Mechanism Method used Author Advantages Disadvantages
1 Neural network
for classification
Probability
estimation for
Posterior
Peter Zhang
Guoqiang
The important
aspects is solving
classification
problems are
discussed
lack of the
rigorous
2 ANN for dataset
classification
Backpropagation
algorithm
Rojalina
Priyadarshini
The best
combination of
training, learning
and transfer
function for
classification of
datasets is found
The combination
doesn’t worked
for larger
datasets
3 Classification of
car seat fabrics
Computer based
tool is used
R.Furferi, L.
Governi
Classified the
fabrics in 93.8%
of the cases
Result varies
from operator to
operator
4 Classification of
IRS-1D satellite
images
Backpropagation
algorithm
E. Hosseini
Aria, J. Amini,
M.R.Saradjian
BPNN is more
accurate
Maximum
likelihood
classification is
less accurate
than BPNN
5 Classification of
neck movement
patterns
Backpropagation
algorithm
Helena Grip,
Fredrik Öhberg
the predictive
ability of a BPNN,
using neck
movement
variables as input
is analysed and it
produced the
result with 90%
accuracy
data
concerning age,
proprioception,
muscle activity
are not discussed
6 Misfire detection
using NN
Feed-forward
multiple layer NN
Abid Ali, Olaf
Magnor
It detects all
patterns of misfire
potential
implementation
factors, like
fixed-point
operations,
quantization are
not discussed
7 Remote sensing
image
classification
Backpropagation
algorithm
Yu-guo Wang,
Hua-peng Li
BP ANN with a
mapping function
is used and the
image
classification is
improved
Not suitable for
wide spectral
distribution
8 Gene Expression
Data
Classification
Samples filtering Wutao
Chen,Huijuan
Lu,Mingyi
Wang
The result is more
accurate and stable
It is unstable for
single neural
network

18
4. CONCULSION
Artificial neural networks (ANN) say that the most active research and application an area is
classification.The neural network was trained by back propagation algorithm. The dataset
classification uses the most effective tool called back propagation neural network. The usage of
Back Propagation Neural Network (BPNN) for classifying the images in remote sensing
technology is also examined. When the maximum likelihood method was compared with
backpropagation neural network method, the BPNN was more accurate than maximum likelihood
method. The classification of neck movement patterns related to Whiplash-associated disorders
(WAD) uses a resilient back propagation neural network (BPNN). The resilient back propagation
neural network (BPNN) resulted in a correct prediction for 84 percent of the control subjects and
89 percent of the WAD, showing that a BPNN could be suitable for predicting motion
characteristics. The statistical analysis method for purifying training samples for remote sensing
classification based on BP ANN is performed. The experiments showed that it was an effective
method to improve image classification. Analyses of breast cells characteristics have great
importance in diagnosis, treatment and following of this disease. In this study, classification of
699 instances of breast cancer data that is available in UCI is performed through artificial neural
network algorithm. Multilayer feed-forward neural network algorithm is also used for
classification, application of misfire detection in gasoline engines, classification of car seat
fabrics etc. however BPNN proves to be more effective than other classification algorithms.
REFERENCES
[1] Rojalina Priyadarshini; “Functional Analysis of Artificial Neural Network for Dataset Classification”,
Special Issue of IJCCT Vol. 1 Issue 2, 3, 4; 2010 for International Conference [ACCTA-2010], 3-5
August 2010.
[2] Guoqiang Peter Zhang “Neural Network for Classification- A Survey 2000” IEEE Transactions on
systems, man and cybermetics- part c: applications and reviews, Vol 30,
[3] R.Furferi, L. Governi; “Neural Network based classification of car seat fabrics” International Journal
of Mathematical Models and Methods in Applied Sciences; Issues 3, Vol. 5, 2011.
[4] E. Hosseini Aria, J. Amini, M.R.Saradjian, “Back Propagation Neural Network for Classification of
IRS-1D Satellite Images” Vol.1, Issue.2, 2003
[5] Helena Grip, Fredrik Öhberg, Urban Wiklund, Ylva Sterner, J. Stefan Karlsson, and Björn Gerdle,
“Classification of Neck Movement Patterns Related to Whiplash-Associated Disorders Using Neural
Networks”, IEEE transactions on information technology in biomedicine, Vol.7, Issue.4,2003.
[6] Yu-guo Wang, Hua-peng Li ,”Remote sensing image classification based on artificial neural network
“,International Conference on Computer, Mechatronics, Control and Electronic Engineering (CMCE)
, Vol.1, Issue.2, 2010.
[7] Guoqiang Peter Zhang, “Classification of Breast Cancer Data with Harmony Search and Back
Propagation Based Artificial Neural Network”, IEEE 22nd Signal Processing and Communications
Applications Conference, 2014.
[8] Guoqiang Peter Zhang , “Neural Networks for Classification: A Survey “,IEEE transactions on
systems, man, and cybernetics , Vol.30, Issue.4, 2000.
[9] Abid Ali, Olaf Magnor and Matthias Schultalbers , “Misfire Detection Using a Neural Network Based
Pattern Recognition “,International Conference on Artificial Intelligence and Computational
Intelligence “, Vol.2, Issue.3, 2009.
[10] R. Furferi, L. Governi and Y. Volpe, “Neural Network based classification of car seat fabrics”,
International journal of mathematical models and methods in applied sciences, Vol.3, Issue.5, 2011.

Review on classification based on artificial

More Related Content

What's hot (19)

Viewers also liked (9)

Similar to Review on classification based on artificial (20)

More from ijasa (20)

Recently uploaded (20)

Review on classification based on artificial